]>
Commit | Line | Data |
---|---|---|
e976a2b9 CB |
1 | /* |
2 | * kvm_virtio.c - virtio for kvm on s390 | |
3 | * | |
4 | * Copyright IBM Corp. 2008 | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License (version 2 only) | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * Author(s): Christian Borntraeger <borntraeger@de.ibm.com> | |
11 | */ | |
12 | ||
13 | #include <linux/init.h> | |
14 | #include <linux/bootmem.h> | |
15 | #include <linux/err.h> | |
16 | #include <linux/virtio.h> | |
17 | #include <linux/virtio_config.h> | |
5a0e3ad6 | 18 | #include <linux/slab.h> |
faeba830 | 19 | #include <linux/virtio_console.h> |
e976a2b9 CB |
20 | #include <linux/interrupt.h> |
21 | #include <linux/virtio_ring.h> | |
17f34580 | 22 | #include <linux/pfn.h> |
e976a2b9 CB |
23 | #include <asm/io.h> |
24 | #include <asm/kvm_para.h> | |
25 | #include <asm/kvm_virtio.h> | |
26 | #include <asm/setup.h> | |
27 | #include <asm/s390_ext.h> | |
28 | ||
29 | #define VIRTIO_SUBCODE_64 0x0D00 | |
30 | ||
31 | /* | |
32 | * The pointer to our (page) of device descriptions. | |
33 | */ | |
34 | static void *kvm_devices; | |
cefa33e2 | 35 | struct work_struct hotplug_work; |
e976a2b9 | 36 | |
e976a2b9 CB |
37 | struct kvm_device { |
38 | struct virtio_device vdev; | |
39 | struct kvm_device_desc *desc; | |
40 | }; | |
41 | ||
42 | #define to_kvmdev(vd) container_of(vd, struct kvm_device, vdev) | |
43 | ||
44 | /* | |
45 | * memory layout: | |
46 | * - kvm_device_descriptor | |
47 | * struct kvm_device_desc | |
48 | * - configuration | |
49 | * struct kvm_vqconfig | |
50 | * - feature bits | |
51 | * - config space | |
52 | */ | |
53 | static struct kvm_vqconfig *kvm_vq_config(const struct kvm_device_desc *desc) | |
54 | { | |
55 | return (struct kvm_vqconfig *)(desc + 1); | |
56 | } | |
57 | ||
58 | static u8 *kvm_vq_features(const struct kvm_device_desc *desc) | |
59 | { | |
60 | return (u8 *)(kvm_vq_config(desc) + desc->num_vq); | |
61 | } | |
62 | ||
63 | static u8 *kvm_vq_configspace(const struct kvm_device_desc *desc) | |
64 | { | |
65 | return kvm_vq_features(desc) + desc->feature_len * 2; | |
66 | } | |
67 | ||
68 | /* | |
69 | * The total size of the config page used by this device (incl. desc) | |
70 | */ | |
71 | static unsigned desc_size(const struct kvm_device_desc *desc) | |
72 | { | |
73 | return sizeof(*desc) | |
74 | + desc->num_vq * sizeof(struct kvm_vqconfig) | |
75 | + desc->feature_len * 2 | |
76 | + desc->config_len; | |
77 | } | |
78 | ||
5ca9fd54 HC |
79 | /* This gets the device's feature bits. */ |
80 | static u32 kvm_get_features(struct virtio_device *vdev) | |
e976a2b9 | 81 | { |
5ca9fd54 HC |
82 | unsigned int i; |
83 | u32 features = 0; | |
e976a2b9 | 84 | struct kvm_device_desc *desc = to_kvmdev(vdev)->desc; |
5ca9fd54 | 85 | u8 *in_features = kvm_vq_features(desc); |
e976a2b9 | 86 | |
5ca9fd54 HC |
87 | for (i = 0; i < min(desc->feature_len * 8, 32); i++) |
88 | if (in_features[i / 8] & (1 << (i % 8))) | |
89 | features |= (1 << i); | |
90 | return features; | |
91 | } | |
e976a2b9 | 92 | |
c624896e | 93 | static void kvm_finalize_features(struct virtio_device *vdev) |
5ca9fd54 | 94 | { |
c624896e | 95 | unsigned int i, bits; |
5ca9fd54 HC |
96 | struct kvm_device_desc *desc = to_kvmdev(vdev)->desc; |
97 | /* Second half of bitmap is features we accept. */ | |
98 | u8 *out_features = kvm_vq_features(desc) + desc->feature_len; | |
e976a2b9 | 99 | |
e34f8725 RR |
100 | /* Give virtio_ring a chance to accept features. */ |
101 | vring_transport_features(vdev); | |
102 | ||
5ca9fd54 | 103 | memset(out_features, 0, desc->feature_len); |
c624896e RR |
104 | bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8; |
105 | for (i = 0; i < bits; i++) { | |
106 | if (test_bit(i, vdev->features)) | |
5ca9fd54 HC |
107 | out_features[i / 8] |= (1 << (i % 8)); |
108 | } | |
e976a2b9 CB |
109 | } |
110 | ||
111 | /* | |
112 | * Reading and writing elements in config space | |
113 | */ | |
114 | static void kvm_get(struct virtio_device *vdev, unsigned int offset, | |
115 | void *buf, unsigned len) | |
116 | { | |
117 | struct kvm_device_desc *desc = to_kvmdev(vdev)->desc; | |
118 | ||
119 | BUG_ON(offset + len > desc->config_len); | |
120 | memcpy(buf, kvm_vq_configspace(desc) + offset, len); | |
121 | } | |
122 | ||
123 | static void kvm_set(struct virtio_device *vdev, unsigned int offset, | |
124 | const void *buf, unsigned len) | |
125 | { | |
126 | struct kvm_device_desc *desc = to_kvmdev(vdev)->desc; | |
127 | ||
128 | BUG_ON(offset + len > desc->config_len); | |
129 | memcpy(kvm_vq_configspace(desc) + offset, buf, len); | |
130 | } | |
131 | ||
132 | /* | |
133 | * The operations to get and set the status word just access | |
134 | * the status field of the device descriptor. set_status will also | |
135 | * make a hypercall to the host, to tell about status changes | |
136 | */ | |
137 | static u8 kvm_get_status(struct virtio_device *vdev) | |
138 | { | |
139 | return to_kvmdev(vdev)->desc->status; | |
140 | } | |
141 | ||
142 | static void kvm_set_status(struct virtio_device *vdev, u8 status) | |
143 | { | |
144 | BUG_ON(!status); | |
145 | to_kvmdev(vdev)->desc->status = status; | |
146 | kvm_hypercall1(KVM_S390_VIRTIO_SET_STATUS, | |
147 | (unsigned long) to_kvmdev(vdev)->desc); | |
148 | } | |
149 | ||
150 | /* | |
151 | * To reset the device, we use the KVM_VIRTIO_RESET hypercall, using the | |
152 | * descriptor address. The Host will zero the status and all the | |
153 | * features. | |
154 | */ | |
155 | static void kvm_reset(struct virtio_device *vdev) | |
156 | { | |
157 | kvm_hypercall1(KVM_S390_VIRTIO_RESET, | |
158 | (unsigned long) to_kvmdev(vdev)->desc); | |
159 | } | |
160 | ||
161 | /* | |
162 | * When the virtio_ring code wants to notify the Host, it calls us here and we | |
163 | * make a hypercall. We hand the address of the virtqueue so the Host | |
164 | * knows which virtqueue we're talking about. | |
165 | */ | |
166 | static void kvm_notify(struct virtqueue *vq) | |
167 | { | |
168 | struct kvm_vqconfig *config = vq->priv; | |
169 | ||
170 | kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, config->address); | |
171 | } | |
172 | ||
173 | /* | |
174 | * This routine finds the first virtqueue described in the configuration of | |
175 | * this device and sets it up. | |
176 | */ | |
177 | static struct virtqueue *kvm_find_vq(struct virtio_device *vdev, | |
9499f5e7 RR |
178 | unsigned index, |
179 | void (*callback)(struct virtqueue *vq), | |
180 | const char *name) | |
e976a2b9 CB |
181 | { |
182 | struct kvm_device *kdev = to_kvmdev(vdev); | |
183 | struct kvm_vqconfig *config; | |
184 | struct virtqueue *vq; | |
185 | int err; | |
186 | ||
187 | if (index >= kdev->desc->num_vq) | |
188 | return ERR_PTR(-ENOENT); | |
189 | ||
190 | config = kvm_vq_config(kdev->desc)+index; | |
191 | ||
17f34580 | 192 | err = vmem_add_mapping(config->address, |
db405988 RR |
193 | vring_size(config->num, |
194 | KVM_S390_VIRTIO_RING_ALIGN)); | |
17f34580 | 195 | if (err) |
e976a2b9 | 196 | goto out; |
e976a2b9 | 197 | |
87c7d57c RR |
198 | vq = vring_new_virtqueue(config->num, KVM_S390_VIRTIO_RING_ALIGN, |
199 | vdev, (void *) config->address, | |
9499f5e7 | 200 | kvm_notify, callback, name); |
e976a2b9 CB |
201 | if (!vq) { |
202 | err = -ENOMEM; | |
203 | goto unmap; | |
204 | } | |
205 | ||
206 | /* | |
207 | * register a callback token | |
208 | * The host will sent this via the external interrupt parameter | |
209 | */ | |
210 | config->token = (u64) vq; | |
211 | ||
212 | vq->priv = config; | |
213 | return vq; | |
214 | unmap: | |
17f34580 | 215 | vmem_remove_mapping(config->address, |
db405988 RR |
216 | vring_size(config->num, |
217 | KVM_S390_VIRTIO_RING_ALIGN)); | |
e976a2b9 CB |
218 | out: |
219 | return ERR_PTR(err); | |
220 | } | |
221 | ||
222 | static void kvm_del_vq(struct virtqueue *vq) | |
223 | { | |
224 | struct kvm_vqconfig *config = vq->priv; | |
225 | ||
226 | vring_del_virtqueue(vq); | |
17f34580 | 227 | vmem_remove_mapping(config->address, |
db405988 RR |
228 | vring_size(config->num, |
229 | KVM_S390_VIRTIO_RING_ALIGN)); | |
e976a2b9 CB |
230 | } |
231 | ||
d2a7ddda MT |
232 | static void kvm_del_vqs(struct virtio_device *vdev) |
233 | { | |
234 | struct virtqueue *vq, *n; | |
235 | ||
236 | list_for_each_entry_safe(vq, n, &vdev->vqs, list) | |
237 | kvm_del_vq(vq); | |
238 | } | |
239 | ||
240 | static int kvm_find_vqs(struct virtio_device *vdev, unsigned nvqs, | |
241 | struct virtqueue *vqs[], | |
242 | vq_callback_t *callbacks[], | |
243 | const char *names[]) | |
244 | { | |
245 | struct kvm_device *kdev = to_kvmdev(vdev); | |
246 | int i; | |
247 | ||
248 | /* We must have this many virtqueues. */ | |
249 | if (nvqs > kdev->desc->num_vq) | |
250 | return -ENOENT; | |
251 | ||
252 | for (i = 0; i < nvqs; ++i) { | |
253 | vqs[i] = kvm_find_vq(vdev, i, callbacks[i], names[i]); | |
254 | if (IS_ERR(vqs[i])) | |
255 | goto error; | |
256 | } | |
257 | return 0; | |
258 | ||
259 | error: | |
260 | kvm_del_vqs(vdev); | |
261 | return PTR_ERR(vqs[i]); | |
262 | } | |
263 | ||
e976a2b9 CB |
264 | /* |
265 | * The config ops structure as defined by virtio config | |
266 | */ | |
267 | static struct virtio_config_ops kvm_vq_configspace_ops = { | |
5ca9fd54 | 268 | .get_features = kvm_get_features, |
c624896e | 269 | .finalize_features = kvm_finalize_features, |
e976a2b9 CB |
270 | .get = kvm_get, |
271 | .set = kvm_set, | |
272 | .get_status = kvm_get_status, | |
273 | .set_status = kvm_set_status, | |
274 | .reset = kvm_reset, | |
d2a7ddda MT |
275 | .find_vqs = kvm_find_vqs, |
276 | .del_vqs = kvm_del_vqs, | |
e976a2b9 CB |
277 | }; |
278 | ||
279 | /* | |
280 | * The root device for the kvm virtio devices. | |
281 | * This makes them appear as /sys/devices/kvm_s390/0,1,2 not /sys/devices/0,1,2. | |
282 | */ | |
37f1c012 | 283 | static struct device *kvm_root; |
e976a2b9 CB |
284 | |
285 | /* | |
286 | * adds a new device and register it with virtio | |
287 | * appropriate drivers are loaded by the device model | |
288 | */ | |
b769f579 | 289 | static void add_kvm_device(struct kvm_device_desc *d, unsigned int offset) |
e976a2b9 CB |
290 | { |
291 | struct kvm_device *kdev; | |
292 | ||
293 | kdev = kzalloc(sizeof(*kdev), GFP_KERNEL); | |
294 | if (!kdev) { | |
b769f579 RR |
295 | printk(KERN_EMERG "Cannot allocate kvm dev %u type %u\n", |
296 | offset, d->type); | |
e976a2b9 CB |
297 | return; |
298 | } | |
299 | ||
37f1c012 | 300 | kdev->vdev.dev.parent = kvm_root; |
e976a2b9 CB |
301 | kdev->vdev.id.device = d->type; |
302 | kdev->vdev.config = &kvm_vq_configspace_ops; | |
303 | kdev->desc = d; | |
304 | ||
305 | if (register_virtio_device(&kdev->vdev) != 0) { | |
b769f579 RR |
306 | printk(KERN_ERR "Failed to register kvm device %u type %u\n", |
307 | offset, d->type); | |
e976a2b9 CB |
308 | kfree(kdev); |
309 | } | |
310 | } | |
311 | ||
312 | /* | |
313 | * scan_devices() simply iterates through the device page. | |
314 | * The type 0 is reserved to mean "end of devices". | |
315 | */ | |
316 | static void scan_devices(void) | |
317 | { | |
318 | unsigned int i; | |
319 | struct kvm_device_desc *d; | |
320 | ||
321 | for (i = 0; i < PAGE_SIZE; i += desc_size(d)) { | |
322 | d = kvm_devices + i; | |
323 | ||
324 | if (d->type == 0) | |
325 | break; | |
326 | ||
b769f579 | 327 | add_kvm_device(d, i); |
e976a2b9 CB |
328 | } |
329 | } | |
330 | ||
cefa33e2 AG |
331 | /* |
332 | * match for a kvm device with a specific desc pointer | |
333 | */ | |
334 | static int match_desc(struct device *dev, void *data) | |
335 | { | |
336 | if ((ulong)to_kvmdev(dev_to_virtio(dev))->desc == (ulong)data) | |
337 | return 1; | |
338 | ||
339 | return 0; | |
340 | } | |
341 | ||
342 | /* | |
343 | * hotplug_device tries to find changes in the device page. | |
344 | */ | |
345 | static void hotplug_devices(struct work_struct *dummy) | |
346 | { | |
347 | unsigned int i; | |
348 | struct kvm_device_desc *d; | |
349 | struct device *dev; | |
350 | ||
351 | for (i = 0; i < PAGE_SIZE; i += desc_size(d)) { | |
352 | d = kvm_devices + i; | |
353 | ||
354 | /* end of list */ | |
355 | if (d->type == 0) | |
356 | break; | |
357 | ||
358 | /* device already exists */ | |
359 | dev = device_find_child(kvm_root, d, match_desc); | |
360 | if (dev) { | |
361 | /* XXX check for hotplug remove */ | |
362 | put_device(dev); | |
363 | continue; | |
364 | } | |
365 | ||
366 | /* new device */ | |
367 | printk(KERN_INFO "Adding new virtio device %p\n", d); | |
368 | add_kvm_device(d, i); | |
369 | } | |
370 | } | |
371 | ||
e976a2b9 CB |
372 | /* |
373 | * we emulate the request_irq behaviour on top of s390 extints | |
374 | */ | |
f6649a7e MS |
375 | static void kvm_extint_handler(unsigned int ext_int_code, |
376 | unsigned int param32, unsigned long param64) | |
e976a2b9 | 377 | { |
be3c5832 CB |
378 | struct virtqueue *vq; |
379 | u16 subcode; | |
fc678d67 | 380 | u32 param; |
e976a2b9 | 381 | |
f6649a7e | 382 | subcode = ext_int_code >> 16; |
e976a2b9 CB |
383 | if ((subcode & 0xff00) != VIRTIO_SUBCODE_64) |
384 | return; | |
385 | ||
be3c5832 | 386 | /* The LSB might be overloaded, we have to mask it */ |
f6649a7e | 387 | vq = (struct virtqueue *)(param64 & ~1UL); |
be3c5832 | 388 | |
fc678d67 | 389 | /* We use ext_params to decide what this interrupt means */ |
f6649a7e | 390 | param = param32 & VIRTIO_PARAM_MASK; |
be3c5832 | 391 | |
fc678d67 AG |
392 | switch (param) { |
393 | case VIRTIO_PARAM_CONFIG_CHANGED: | |
394 | { | |
be3c5832 CB |
395 | struct virtio_driver *drv; |
396 | drv = container_of(vq->vdev->dev.driver, | |
397 | struct virtio_driver, driver); | |
398 | if (drv->config_changed) | |
399 | drv->config_changed(vq->vdev); | |
fc678d67 AG |
400 | |
401 | break; | |
402 | } | |
cefa33e2 AG |
403 | case VIRTIO_PARAM_DEV_ADD: |
404 | schedule_work(&hotplug_work); | |
405 | break; | |
fc678d67 AG |
406 | case VIRTIO_PARAM_VRING_INTERRUPT: |
407 | default: | |
be3c5832 | 408 | vring_interrupt(0, vq); |
fc678d67 AG |
409 | break; |
410 | } | |
e976a2b9 CB |
411 | } |
412 | ||
413 | /* | |
414 | * Init function for virtio | |
415 | * devices are in a single page above top of "normal" mem | |
416 | */ | |
417 | static int __init kvm_devices_init(void) | |
418 | { | |
419 | int rc; | |
420 | ||
421 | if (!MACHINE_IS_KVM) | |
422 | return -ENODEV; | |
423 | ||
035da16f | 424 | kvm_root = root_device_register("kvm_s390"); |
37f1c012 CH |
425 | if (IS_ERR(kvm_root)) { |
426 | rc = PTR_ERR(kvm_root); | |
e976a2b9 CB |
427 | printk(KERN_ERR "Could not register kvm_s390 root device"); |
428 | return rc; | |
429 | } | |
430 | ||
cc835f78 | 431 | rc = vmem_add_mapping(real_memory_size, PAGE_SIZE); |
17f34580 | 432 | if (rc) { |
035da16f | 433 | root_device_unregister(kvm_root); |
17f34580 | 434 | return rc; |
e976a2b9 CB |
435 | } |
436 | ||
cc835f78 | 437 | kvm_devices = (void *) real_memory_size; |
e976a2b9 | 438 | |
cefa33e2 AG |
439 | INIT_WORK(&hotplug_work, hotplug_devices); |
440 | ||
e976a2b9 CB |
441 | ctl_set_bit(0, 9); |
442 | register_external_interrupt(0x2603, kvm_extint_handler); | |
443 | ||
444 | scan_devices(); | |
445 | return 0; | |
446 | } | |
447 | ||
faeba830 CB |
448 | /* code for early console output with virtio_console */ |
449 | static __init int early_put_chars(u32 vtermno, const char *buf, int count) | |
450 | { | |
451 | char scratch[17]; | |
452 | unsigned int len = count; | |
453 | ||
454 | if (len > sizeof(scratch) - 1) | |
455 | len = sizeof(scratch) - 1; | |
456 | scratch[len] = '\0'; | |
457 | memcpy(scratch, buf, len); | |
458 | kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, __pa(scratch)); | |
459 | return len; | |
460 | } | |
461 | ||
c4de0c1a | 462 | static int __init s390_virtio_console_init(void) |
faeba830 | 463 | { |
c4de0c1a HB |
464 | if (!MACHINE_IS_KVM) |
465 | return -ENODEV; | |
466 | return virtio_cons_early_init(early_put_chars); | |
faeba830 | 467 | } |
c4de0c1a HB |
468 | console_initcall(s390_virtio_console_init); |
469 | ||
faeba830 | 470 | |
e976a2b9 CB |
471 | /* |
472 | * We do this after core stuff, but before the drivers. | |
473 | */ | |
474 | postcore_initcall(kvm_devices_init); |